Accelerating Belief Propagation with Task-Based Hardware Parallelism

BASc Thesis, Supervised by Professor Mark Jeffrey | University of Toronto | Toronto, Canada | Apr 2024

This project is the convergence of two directions, the first being innovations in residual belief propagation. Belief propagation is an algorithm used to compute statistical inferences on graphs called probabilistic graphical models, where random variables form nodes, and a probability mass function forms edges. Belief propagation takes these models and computes the marginal distributions of all the random variable nodes. Applications include stereo image depth estimation, workplace safety, and healthcare patient experience. Existing implementations of BP fail to achieve acceptable convergence coverage, convergence rate, and linear scaling with hardware resources. The second direction is hardware support for priority-ordered algorithms through task-based parallelism [10]. Speculative hardware parallel execution of BP has been simulated and implemented with CPUs, but no cost-effective pure hardware implementation exists. This work implements belief propagation on an FPGA-based speculative parallel accelerator and demonstrates the possibility of increased performance.
Accelerating Belief Propagation with Task-Based Hardware Parallelism

Automation of Thermal Energy Storage for Homes using Artificial Neural Networks

IEEE Canadian Conference on Electrical and Computer Engineering | London, Canada | Sep 2020

About 60% of the energy consumed by homes in North America is for air conditioning. With about 78% of electric energy is generated by from fossil fuels in the US, this energy use contributes to greenhouse gas emissions and global warming. Residential solar energy is now becoming cost effective and is as cost effective electric energy from the electric grid. However, solar energy availability and energy required for air conditioning are mismatched with respect to time. This mismatch in availability and need necessitates the use of energy storage. In previous works, storage of energy in thermal air mass of homes has been proposed. However, the thermostat required for such application is very complex. In this work, an artificial-neural-network-based thermostat is proposed. A method to train the model for an average home is demonstrated with an example, and the method is shown to be effective.
DOI: 10.1109/CCECE47787.2020.9255680

Thermal Energy Storage for Homes

2018 IEEE International Conference on Smart Energy Grid Engineering | Oshawa, Canada | Aug 2018

Conventional energy management solutions for homes that use sun energy typically convert energy from the sun, using photovoltaic (PV) panels, and then stores the energy into batteries. However, the use of batteries is not environmentally friendly and is expensive. They become a limitation in the effective use of sun energy. This paper proposes a solution based on thermal energy storage. Thermal Energy Storage for Homes (TESH) is a solution to the problem of mismatched timing of solar energy and home energy demand when using solar power. By using the air mass in the home to store thermal energy, through by altering its temperature, one can avoid the need for other forms of energy storage, such as batteries. Batteries add to costs and pose environmental hazards, while the use of air mass as a thermal energy storage medium is cost-free. This makes solar energy an even more environmentally friendly alternative energy source, while simultaneously reducing the cost of infrastructure. This will allow more people to purchase and utilize solar energy, while also having less of a negative impact on the environment. The proposed method was implemented as a prototype and tested. Test results are reported and discussed.
DOI: 10.1109/SEGE.2018.8499511


Google Scholar